Adaptive Dynamics Planning for Robot Navigation

TL;DR

This work introduces Adaptive Dynamics Planning (ADP), a learning-augmented framework that dynamically adjusts robot dynamics fidelity during trajectory planning to balance accuracy and computational cost. By formulating ADP as an MDP and optimizing a TD3-based policy, the approach tunes dynamics configuration in real time according to environmental cues, and integrates it with three classical planners (DWA, MPPI, Log-MPPI) as well as a standalone navigation system. Extensive simulations on the BARN benchmark and real-world Jackal experiments show that ADP consistently improves navigation success, safety, and efficiency, particularly in highly constrained or complex environments. The results demonstrate strong generalization to unseen environments and highlight the pragmatic value of environment-aware dynamics modeling for real-time robotic navigation.

Abstract

Autonomous robot navigation systems often rely on hierarchical planning, where global planners compute collision-free paths without considering dynamics, and local planners enforce dynamics constraints to produce executable commands. This discontinuity in dynamics often leads to trajectory tracking failure in highly constrained environments. Recent approaches integrate dynamics within the entire planning process by gradually decreasing its fidelity, e.g., increasing integration steps and reducing collision checking resolution, for real-time planning efficiency. However, they assume that the fidelity of the dynamics should decrease according to a manually designed scheme. Such static settings fail to adapt to environmental complexity variations, resulting in computational overhead in simple environments or insufficient dynamics consideration in obstacle-rich scenarios. To overcome this limitation, we propose Adaptive Dynamics Planning (ADP), a learning-augmented paradigm that uses reinforcement learning to dynamically adjust robot dynamics properties, enabling planners to adapt across diverse environments. We integrate ADP into three different planners and further design a standalone ADP-based navigation system, benchmarking them against other baselines. Experiments in both simulation and real-world tests show that ADP consistently improves navigation success, safety, and efficiency.

Figures (4)

• Figure 1: Instead of using a fixed scheme to decrease dynamics fidelity as adopted by DDP lu2025decremental (top), ADP leverages Reinforcement Learning (RL) to adaptively adjust dynamics fidelity to balance modeling accuracy and computation efficiency during trajectory rollout (bottom).
• Figure 2: Example of the ADP Navigation System Operating in a BARN Environment. Left: Gazebo Visualization. Right: RViz Visualization. Lighter and deeper green colors indicate lower and higher fidelity in dynamics modeling, respectively.
• Figure 3: Two Physical Test Environments.
• Figure 4: Real-World Cluttered Environment Experiments.